Social dual-effect driven group modeling for neural group recommendation

Frequent group activities of human beings have become an indispensable part of human daily life. Group recommendation aims to recommend preferred items to a group of users in recommender systems. The existing solutions on group recommendation are to explore group modeling with or without the help of auxiliary information on social networks. However, we observe that the social factors can be explored directly in the group without employing social information on social networks. Towards this end, we study the social effect-based design guideline to drive group modeling. In this work, we propose a novel Social dual-Effect driven Attentive Group Recommendation method (SEAGR) that well utilizes social selection effect and social influence effect from sociology to explore group representation learning for neural group recommendation. Specifically, we construct the social selection-driven group inherent modeling from interaction-level and user-level. To mimic interaction-based dynamic group decision-making, we also design a social influence-driven attentive influence mining model in terms of users’ influence distinction in different groups. Based on these two components, an aggregative group representation is obtained. Moreover, neural recommendation for groups and users could be intensified reciprocally considering the impact of groups on users. The experimental results validate the effectiveness of the proposed method on three real-world datasets, and demonstrate its advantages over state-of-the-art methods in accuracy through extensive experiments.